probability modeling and simulation of insurance claims in ...€¦ · validate the desired...
TRANSCRIPT
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
41
Probability Modeling and Simulation of Insurance Claims in Ghana
Godfred Kwame Abledu1, Elvis Dadey & Agbodah Kobina
1. Coresponding author: Applied Mathematics Deprtment, Koforidua Polytecnic-Ghana
ABSTRACT
Probability modeling has a wide range of applications in the field of insurance and finance. Over the years, its
applications have been modified to include simulation techniques which in most cases are engaged to enhanced or
validate the desired estimates. This study investigates the probability distributions that best fit a number of insurance
claims. In particular, it compares the Poisson distribution and the negative binomial distribution models to determine
which distribution best fits insurance claim data obtained from two Insurance Companies in Ghana. Data on number of claims of a funeral policy spanning from 2006 to 2010 were used for the study. Probability
distribution models and the parametric bootstrap methods were employed in analyzing the data collected. The result of
the study revealed that on the average, the number of claims tends to increase as the year go by. Also the Negative
Binomial Distribution was found to be superior to the Poisson distribution in fitting the claims data. Finally, a
comparison between the estimates obtained by the probability models and that of the parametric bootstrap estimates
revealed no significant difference.
Keywords: parametric bootstrap, insurance claims, simulation techniques, negative binomial distribution, Poisson
distribution.
Introduction The risk involved in allocating insurance funds has boosted the need for statistical methods and principles.
Insurance funds may be invested in several asset categories, which include bonds, equities and others with the aim of
increasing investment real returns to meet claim payment demands and other financial obligations.
The puzzle of percentage allocation of insurance fund into investment portfolios is subject to insurance liabilities of
which part is systematic and the other stochastic. Insurance liability relating to stochastic constraints, which is of much
concern to this research work forms comparatively huge part of the total liability of insurance companies. It is solely
linked to claim demands and uncertainties relating to it by which the policy contract mandates the insurer to forfeit. Insurance companies employ statistical methods of estimation to acquire concrete information about these uncertain
liabilities with which decisions pertaining to assets allocation, expected monthly claims and payment targets as well as
future insurance pricing can be ascertained. This is known in insurance administration as the expected claim liability
estimation problem.
The roots of insurance might be traced to Babylonia, where traders were encouraged to assume the risks of the
caravan trade through loans with interest that were rapid only after the goods had arrived safely. The Phoenicians and the
Greeks applied a similar system to their seaborne commerce. The Romans used burial clubs as a form of life insurance,
providing funeral expenses for members and later payments to the survivors.
(http://www.infoplease.com/business/insurance)
Insurance developed rapidly with the growth of British commerce in the 17th and 18th centuries. Prior to the
formation of corporations devoted solely to the business of writing insurance, policies were signed by a number of individuals, each of whom wrote his name and the amount of risk he was assuming underneath the insurance proposal.
(http://encyclopedia2.thefreedictionary.com/Insurance).
The first stock companies to engage in insurance were chartered in England in 1720 and 1735. Fire insurance
corporations were formed in New York City in 1787 and at Philadelphia in 1794. The Presbyterian synod of Philadelphia
sponsored the first life insurance corporation in America for the benefits of Presbyterian ministers and their dependants.
After 1840, with the decline of religious prejudice against the practice, life insurance entered a boom period. In the 1830s
the practice of classifying risks began. (http://www.infoplease.com/encyclopedia/business/insurance).
In the 19th century, many friendly societies were founded to insure the life and health of their members. Similarly,
fraternal orders and most labour organizations continued to provide insurance coverage for its members. Currently, many
employers sponsor Group Insurance Policies for their employees. Such policies generally include not only life insurance
but sickness, accident benefits and old-age pensions. The employees usually contribute a certain percentage of the
premium. Since the 19th century, there has been a growing tendency for the state to enter the field of insurance, especially with respect to safeguarding workers against sickness and disability, either temporary or permanent, destitute
old age and unemployment.
(http://www.infoplease.com/encyclopedia/business/insurance).
Classification Insurance policy may be classified into two types namely, Life Insurance policy and Non-Life Insurance policy.
Life insurance policy is originally conceived to protect a man's family when his death left them without income. This has
developed into variety of policy plans. For example, in a Whole Life insurance policy, fixed premiums are paid
throughout the insured's life time. This accumulated amount, augmented by compound interests, is paid to a beneficiary
in a lump sum upon the insured's death. The benefits are paid even if the insured had terminated the policy.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
42
Universal life insurance was created to provide more flexibility than whole life insurance by allowing the policy
owner to shift money between the insurance and savings components of the policy. Premiums, which are variable, are
broken down by the insurance company into insurance and savings, allowing the policy owner to make adjustments
based on their individual circumstances. For example, if the savings portion is earning a low return, it can be used instead
of external funds to pay the premiums. Unlike whole life insurance, universal life allows the cash value of investments to
grow at a variable rate that is adjusted monthly.
(http://www.investopedia.com/terms/u/universallife.asp). A typical example of a Non Life Insurance policy is Fire Insurance. This includes damage from lightning,
tornado, flood and drought. Complete automobile insurance includes not only insurance against fire and theft but also
against compensation for damage to a car and for personal injury to the victim of an accident (liability insurance); many
car owners; however carry only partial insurance (http://encyclopedia2.thefreedictionary.com/Insurance).
In many countries, liability insurance is compulsory. Bonding or fidelity insurance is designed to protect an
employer against dishonesty on the part of an employee. Title insurance is aimed at protecting purchasers of real estate
from loss by reason of defective title. Credit insurance safeguards business against loss from the failure of customers to
meet their obligations. Marine insurance protects shipping companies against the loss of a ship or its cargo, as well as
many other items(http://encyclopedia2.thefreedictionary.com/Insurancee).
In recent years, the insurance industry has broadened to guard against almost any conceivable risk; companies
will insure a dancer's legs, a pianist fingers or an outdoor event against loss from rain on a specified day. A hybrid type
of insurance is what we call composite insurance policy, which combines both life and non life insurance.
Insurance in Ghana Insurance in Ghana began in the colonial era,1924, with the establishment of the Royal Guardian Enterprise now
known as the Enterprise Insurance Company Limited. The first indigenous private insurance company, the Gold Coast
Insurance Company was established in 1955. In 1962, the State Insurance Company of Ghana (SIC) was set up. Eleven
(11) more companies were established in 1971. Five years later seven (7) more insurance companies, one (1) reinsurance industry and insurance brokerage firms were established.
The 1989 insurance law, PNDC Law 229 led to the establishment of the National Insurance Commission. The
National Insurance Commission (NIC) remains the sole institution that has been mandated to regulate and supervise
insurance activities within the country. The insurance industry is thriving so much that the business is quite common
across the globe. The major insurance companies in Ghana can be grouped into:(i) Life Insurance, (ii) Non-Life
Insurance, and (iii) Composite Insurance (a combination of Life and Non-Life insurance).
The insurance industry as of 2010 statistics was made up of the following: 23 Non-Life companies, 17 Life
companies, 2 Reinsurance companies with over 40 Broking companies, one reinsurance broking company, one loss
adjusting company and about 4,000 insurance agents.
In Ghana, insurance is an accepted thing to do, as everyone must protect themselves against risks in life. The needs
of insurance in Ghana for example cuts across travel insurance, fire insurance, motor insurance, marine insurance etc.
Though quite competitive, the governing laws of insurance have also increased with time. For example, in Ghana, Insurance Act 58 stipulates that all car owners insure their vehicles against unintended risks and damages. Despite the
many posing questions asked on insurance, Ghanaians are gradually accepting the fact of paying premiums to protect
themselves and their loved ones in case of loss and damage.
This can be attributed to the rude awakening of the Great Fire of London in 1966 destroying over 13,200 houses and
lives as well as other Ghanaians properties (houses and market buildings) that were damaged by fire and flood in the
early years without the owners insuring them. To a greater extent the problems faced by the insurance industry lies in risk
management in the wake of financial crisis, that is maintaining adequate capital and surplus in support of continuous
operations. Poor public perception of insurance has also resulted in lack of confidence in insurance. Again changes in
sales and distribution as a result of decline in business volumes and changes in customer focus and related operating
modules posses as a threat to the industry. Ineffective international financial reporting standards and ineffective
investment management procedures are just a few to highlight within the industry. (http://mybusinessweekafrica.com/topheadlines-detail.php).
Statement of the Problem The statistics on industries in Ghana revealed that insurance companies have increased appreciably over the last five
(5) years and the demand of Ghanaians to secure insurance policies for their business and households is overwhelming
(http://www.nicgh.org/live/en/?pg=104.pp). To this end, Insurance Companies are repositioning themselves to meet the
future challenges of demand for Insurance Policies, Risk Management and Investments as well as Claim Payment Demands. In view of the fact that every policy holder expects a cushion in the event of his/her economic loss as
stipulated in an Insurance Contract, the challenge of meeting the payment terms becomes an issue of much concern to the
Insurer.
The estimation of Expected Claim Liabilities of the Insurance Policies enables decisions on asset allocation and
claim payment to be taken without much error thereby rising above the said challenges. The aim of this research is to
explore probability models to model a number of Insurance Claims and to estimate the expected claim Liabilities of
Insurance Policies.
Research Objectives The main objective of the study was to explore probability models that would model adequately the number of
claims occurring under funeral insurance policies in Ghana to estimate the expected number of claims.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
43
The specific objectives of the study are to:
(i) Identify and explain seasonal variation within the number of claims.
(ii) Compare the Probability distribution estimates of the number of claims.
(iii) Derive the Probability distribution model that best fits the number of Claims for the funeral policies.
(iv) Construct bootstrap confidence intervals for the Expected Number of Claims and compare with that of the estimates
obtained from the models.
Literature Review Probability models of count data provide essential information for Insurance Pricing, Insurance Liability Estimation
and Investment Decision Making. Claims and unexpired risks provisions in insurance companies are determined based
on previous claims experience, knowledge of the events and the terms and conditions of the relevant policies and on
interpretation of circumstances.
Most insurance companies set claim payment targets by considering the estimates obtained by predicting the
number of claims for some unforeseen periods. Insurance claim liability estimation is a sub problem of insurance liability estimation. Most insurance companies prefer to tackle the problem holistically, but all the same the claim liability is
estimated somehow. An example is the Bornhuetter- Ferguson loss reserve method of insurance liability estimation (
Asmussen and Kliippelberg, 2008)
The product of the expected number of claim and the benefit paid under the policy is essentially the expected claim
liability when the policy is a fixed benefit policy and all economic factors such as inflation, interest rates etc are all held
constant. It is noteworthy that a good estimate of the probability distribution model may lead to a good estimate of the
claim liability that will improve on other estimates and the overall insurance liability estimate.
In statistics, stochastic count data refers to data in which the observations are uncertain and can take only non-
negative integer values, i.e. {0, 1, 2, 3,…}. These integers arise from counting rather than ranking and that the statistical
treatment of such data is distinct from that of binary data, in which the observations can take only two values, usually
represented by 0 and 1. The methodology of stochastic count data modeling also known as Probability Modeling evolved from just an assumption and parameter estimation to modeling and has grown considerably in the past years with a lot of
new procedures and methods. This evolution has taken place in both discrete and continuous probability modeling in
diverse field of the sciences with most effects felt in the Statistics and Engineering Sciences (Yaacob et al., 2010).
An appreciable amount of work has being done, making up theories and methods that have added to the modeling
body of knowledge. The theories and methods invented dealt proficiently with the complexities and have rendered the
resolutions mechanic. The Poisson distribution model of stochastic count data requires strong independence assumptions
with regards to the underlining stochastic process and any violation of these assumptions invalidates the model. One
important features of the Poisson distribution that has justified its inadequacy in stochastic count data modeling is the
variance-to-mean ratio, also called Fisher dispersion index, which is equal to one (1) whatever the value of l . This
inappropriateness of modeling count data with the Poisson distribution has long been recognized, and collectively risk
algorithms have being developed that works just as well as other frequency distributions, in particular the Negative
Binomial Distribution (Asmussen, and Kliippelberg, 2008; Mikosch and Nagaev,2007).
In this regards various test for detecting over dispersion have also been developed. Again more general models have
been proposed that allow for over dispersion. Furthermore, its effect has been analyzed and corrected for within the maintained Poisson model (Schlegel, 1998). Other research works(Terce o, and Gil, 2012; Osgood, 2000) have ascribed
this inappropriateness to the under dispersion and over dispersion in count data with the later being the specific case of
the count data for insurance claims. The authors consequently defined over dispersion with respect to insurance claims
modeling as the case where there is evidence that the observed random variation is greater than the expected random
variation under the Poison model while under dispersion means that the expected variation is greater than the observed
variation.
Konstantinides et al., (2002 ) compared the adequacies of the Poisson model to each of the mixed Poisson models as
well as the mixed Poisson models with the Belgian motor third party insurance portfolio observed during the year 1997
and concluded that the mixed Poisson models fits better than the Poisson model. Again the comparison amongst the
mixed Poisson models, which include the Poisson-Normal, Poisson-Lognormal and the Poisson-Inverse Gaussian model
revealed no significant difference.
Mikosch, and Nagaev (2007) in their quest to model count data employed the negative binomial distribution resulting from Poisson-Gamma mixture to investigate small sample properties of the likelihood-based test and compared
their performances to those of the Wilcoxon and t-test. The authors established the inefficiencies of the Poisson model
and illustrated procedures used to compute power and sample sizes to design studies with response variables that are over
dispersed.
Konstantinides et al.,(2002) and Hall and Yao (2003). modeled crush count data with the object to propose an
approach to compare the techniques of count data modeling. The authors proposed a simulation approach that used
Negative Binomial Regression Mean Estimate (NB), Negative Binomial Empirical Bayesian Estimate (NBEB) and
Pearson regression residuals to rank sites and to identify hot sports. The results deduced showed that a small fraction of
the artificial site can be successfully identified regardless of the methodology and Pearson regression led to the best
results.
Alternative model expressions in the family of the Mixed Poisson models were proposed by Asmussen and Albrecher (2010) by introducing an additive and multiplicative random effect Poisson model. The author modeled
injected random effects with the Gamma and Inverse Gaussian distributions, which yielded both univariate and
multivariate models in a form of mathematical expressions. The idea was that count data often exhibit dispersion and
unobserved heterogeneity, which renders the usual Poisson model ineffective thereby soliciting for other appropriate
models. This idea was also evident in the research paper by Asmussen and Kliippelberg(1996) when they intended to fit a
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
44
distribution to a data set on claim size in order to obtain rough estimate of risk premiums.
Random effect models, which provided the best fits were recommended by Ridout et al.(2001) and Migue and
Olave (1999b) in their investigation into models of time dependent insurance claim count based on Generalized Poisson
and Negative Binomial distributions. The recommendations were based on model fitting and comparisons by statistical
specification test for nested and non nested models. Further deductions suggested that these random effect models can be
applied on other claim count distributions such as Hurdle or Zero-inflated model that provided good fits for cross-
sectional data. Yue(2003),Thombs and Schucany (1990) used a nonparametric bootstrap method to obtain the prediction interval of
the autoregressive model. Migue and Olave (1999a) applied this method to the ARCH model. Rodriguez and Ruiz (2008)
proposed using the bootstrap method for state space models, to evaluate estimation errors. However, while developing
prediction intervals, there were usually two limitations: (1) how the uncertainty of parametric estimation can be
combined for the prediction interval and (2) the assumption of error may not be accurate.
To overcome these limitations, Pascual et al. (2006) proposed that, for the GARCH model, the bootstrap method
used to create the prediction interval to combine with the uncertainty of parametric estimation and assumptions that are
independent of te . However, when te had heavy-tailed distribution, this study expanded the method of Pascual et
al.,(2006) and added the concept of the sub-sample bootstrap method proposed by Hall and Yao (2003) and created the
prediction interval in the GARCH model that had heavy-tailed distribution. Lee (2009) applied the bootstrap strategy
proposed by Pascual et al. (2006) to study listed daily stock returns in Taiwan. The results showed that the volatility
intervals created using the bootstrap method could appropriately manage the changes of volatility returns.
These and other studies are the many studies conducted in the field of probability modeling. Today, the results obtained from stochastic data modeling are rarely applied without some simulation due to unidentified stochastic shocks.
In this regards, results obtained from these models are subjected to Monte Carlo simulation comprising bootstrapping,
jackknifing etc in order to further perceive the results for adequate industrial decision to be made.
Methodology The study focused on estimating the expected number of claims under funeral insurance policies sold by two
insurance companies namely StarLife and MetLife insurance company. Data for the study included monthly number of claims filed under the policies spanning for a period of sixty (60) months (January, 2006 to December, 2010). Mathworks
Inc.'s Technical programming Language MatLab, version 7.10.0.499 (R2010a) and the SAS Institute Inc.'s software,
version 1.0.0.80 were the main statistical softwares used to perform the analyses. Microsoft Office Excel 2007 and R
software version 2.13.0 were used to organize the data and produce graphs for the analysis. The final report was written
using the Jonathan Kew, Stefan Loffer version 0.4.0 r. 759 (MikTeX) of TeXworks and TeXnicCenter version 1 Beta
7.50. It is a simple environment for editing, typesetting and previewing TeX documents.
Descriptive statistics of the data were generated to study the distribution of the number of claims for the entire
period and on annual basis. Shapiro-Wilk test of normality and two sided Kolmogorov-Smirnov test were performed to
draw conclusion on the nature of the distributions of the claims. Poisson probability distribution (PoisDis) and Negative
Binomial distribution (NegBin) were used to fit the number of claims and the distribution which produced a better fit was
subjected to an asymptotic chi-square test as a confirmatory analysis. Finally the estimated distribution was simulated to obtain 100 re-samples of artificial data sets for monthly number
of claims and the desired statistics recalculated. This was followed by comparative analysis to address the study
objectives.
Data Analysis The data used for the research were solely secondary data obtained from Star Life and Metropolitan life Insurance
Companies. The data consisted of monthly recorded number of claims under a Family Funeral Insurance Policy for a period of Five years (2006 - 2010). The analyses were conducted in two folds, which included preliminary analyses and
some further detailed analyses.
Exploratory and Confirmatory data analyses were the main techniques employed in the preliminary analysis. First
of all, the number of claims of both StarLife and MetLife insurance companies were summarized with relevant summary
statistics and compared to observe the nature of similarities or otherwise. The number of claims were analyzed on annual
basis to measure the seasonal variations or otherwise that existed within and between the number of claims for both
portfolios.
Again, visual representations in the form of Bar Graph, Histogram, Box Plots, QQ Plots and Density Plots were
inspected in other to compare the underlining distribution of the number of claims to the empirical normal distribution.
Finally, the results of the preliminary analysis were confirmed accordingly with the Shapiro Wilk test of normality and
two sided Kolmogorov-smirnov test of similarity of distribution. The Shapiro-Wilk test was used to test the null hypothesis that a sample x1, x2, x3,… came from a normally distributed population.
Results and Discussions The analysis focused on modeling the number of claims under the funeral policies sold by StarLife and MetLife
insurance companies. The paramount purpose was to fit an appropriate probability distribution model to the sixty (60)
data points of monthly recorded claims observed from the beginning of 2006 to the end of 2010 in order to estimate the
expected number of claims to aid investment allocation. The Poisson and Negative binomial distributions are considered for modeling whereas the simulation method
employed was bootstrapping. The Pearson chi square test together with the scaled deviance test informed the adequacy of
the models.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
45
Summary Statistics
Figure 1 presents a bar plot of the number of claims for MetLife and StarLife Insurance companies on funeral
policy for the period (2006 - 2010). A glance at Figure 1 reveals heavy-tailedness and skewness to the right of number of
Claims for both companies. This distribution is typical of number of insurance claims. The number of claims fluctuate
increasingly within the period for both claims with Metlife experiencing claims between zero (no claim) and sixty-four
(64) claims and StarLife between zero (no claims) and ninety 90 claims. The mean number of claims for MetLife and StarLife over the period (2006 - 2010) was seventeen (17) and thirteen (13) claims respectively (see Table 1.
Figure 1: Bar plot of Number of Claims on Funeral Policy from StarLife and MetLife.
Table 1: Summary Statistics of Number of Claims of Funeral Policy from Starlife and MetLife.
Year StarLife MetLife
Mean 13.05 16.65
Median 7.00 11.50
SD 16.82 15.98
Skewness 2.07 0.89
Kurtosis 6.11 0.07
Seasonal Analysis
A study of the seasonal changes in the occurrence of the number of claims revealed an increasing pattern for both
Portfolios along the period. The highest number of claims was observed in July and March, 2010 for StarLife and
MetLife respectively. Several months between 2006 and 2007 recorded no claims (zero claims), that was linked to the
reason that the policy was introduced in 2005 and as at the end of 2006, not much of the policies had being sold.
Figure 2: Boxplot of Yearly Number of Claims on Funeral Policy from StarLife and MetLife.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
46
Figure 2 is a box plot of the monthly recorded number of claims on yearly basis for the underlying years (2006 -
2010). The yearly distributions of the number of claims were all tailed and skewed to either the left or right in the various
years. The annual distributions of number of claims for MetLife were skewed to the right while those of StarLife were
skewed to the left in 2010 but in the other years under review they were all skewed to the right (2006 - 2009). This
revealed that the annual distributions were not conventionally bell shaped (normally distributed) in any of the years but
were all tailed and skewed.
Considering, the annual distributions during the active years (2008 - 2010) where much of the policies had been sold, the year 2008 distribution of the number of claims from MetLife was heavily tailed, right skewed and flat arched
(platykurtic) while that of StarLife was slightly tailed, right skewed and slender arched (leptokurtic). Furthermore, year
2009 distribution depicted a tailed, right skewed and slender arched (leptokurtic) for MetLife and tailed, right skewed and
normally peaked (mesokurtic) for StarLife.
Finally in year 2010, the distributions for both portfolios were similar, tailed, asymmetric and slender arched
(leptokurtic). Overall, the number of claims depicted increasing patterns across the years with rising annual averages (see
Table 2).
Table 2: Trend of Annual Averages of Number of Claims from Starlife and MetLife
Year 2006 2007 2008 2009 2010
StarLife 0.33 2.08 8.25 17.72 36.67
MetLife 0.33 5.33 22.18 15.83 39.53
Comparison of the Distributions
The preliminary analysis on the number of claims suggests that the distributions of both portfolios were not normal.
The null hypothesis of normality was rejected at a p-value of 81.28 10-´ for StarLife and
81.28 10-´ for MetLife
according to Shapiro-Wilk test of normality.
To answer the question as to whether the distribution of number of claims for both insurance companies came from
the same distribution, a QQ plot displayed in Figure 3 was constructed and the fairly linear trend preliminarily suggest
that the distribution of the number of claims for both insurance companies belong to the same distribution. Moreover, the
p-value of 12.13 10-´ obtained by the two sided Kolmogorov-Smirnov test was statistically significant to ascertain that
the probability distribution of the number of claim were the same.
Figure 3: Quartile plot of Number of Claims on Funeral Policy from StarLife and MetLife.
Fitting Probability Distribution to Claims
The density estimates as displayed in Figure 4 for both portfolios are not significantly different. They both have a
similar distribution but the StarLife distribution curve is stepper and slender than that of MetLife Insurance Company.
Furthermore, it is visible that the sample distributions though unimodal, has several turning points which is not typical of
the conventional probability distribution at a glance. However, ascribing this abnormality to the presence of outliers in
the data set may warrant smoothening (fitting) to depict a true probability distribution. Empirical Distributions that can be fitted to the observed data included the Mixed Poisson Distributions starting with the Poisson distribution as discussed
in the literature.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
47
Figure 4: Density Estimate of Number of Claims on Funeral Policy from StarLife and MetLife.
Figures 5 and 6 were produced as a result of fitting the number of Claims for StarLife and MetLife respectively
with the negative binomial and Poisson distribution. A critical look at the charts (Figures 5 and 6) reveals that the Poisson distribution does not fit well to the number of claims from both StarLife and MetLife Insurance Companies. However the
Negative Binomial Distribution fits the Number of Claims reasonably well.
Figure 5: Negative Binomial and Poisson Distribution Fit to the Number of Claims on Funeral Policy from
StarLife and MetLife.
The maximum likelihood estimates of the Poisson mean for StarLife and MetLife were 13.05sl = and
16.06ml = respectively. The 95% confidence interval for sl and ml were (12:1672; 13:9968) and (15:6489; 17:7152)
and finally, the loglikelihood were 1228:3611 and 1810:5978. However, for the negative binomial distribution, the means
were estimated to be 13.0500sl = and 16.6500sl = and the dispersion parameters to be as = 2.0166 and am =
1.4323 for the StarLife and MetLife number of claims.
Figure 6: Percentile Plot of the Distribution of Number of Claims, Negative Binomial and Poisson distribution.
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
48
The variance of the random effect for both StarLife and MetLife were estimated to be ( )V 0.49588s Q = and
( )V 0.69818m Q = respectively. The 95% confidence interval for the dispersion parameters were (1.2633; 2.7699),
(0.8912; 1.9734) and (0.3610; 0.7916), (0.5067; 1.1221) for their variance ( )V .Q
The Log-likelihood of the Negative Binomial Distribution was 1667.5653 and 2175.5353 for StarLife and MetLife
respectively which were far better than that of the Poisson distribution for both StarLife and MetLife respectively.
Finally, the confidence interval estimates of the monthly expected number of claims estimated by the Negative Binomial
Model were approximately (9; 19) and (12; 23) claims for StarLife and MetLife respectively(see Table 3).
Table 3: Interval Estimates of the Number of Claims from StarLife and MetLife
Company Poisson Estimate Negative
Binomial Estimate
StarLife 12.1672 13.9968 9.0494 18.8196
MetLife 16.6500 16.7274 0.0774 16.5726
Table 4 provides summary statistics of the Goodness of fit of the negative binomial distribution for both StarLife
and MetLife Insurance Companies. The scaled deviance of 69.3796 and 71.2286 for StarLife and MetLife compared to
the asymptotic chi-square with 59 degrees of freedom yielded a p-value of about 0.15, which implies we cannot reject the
null hypothesis that the specified negative binomial model is the correct model. Again, the dispersion parameter
estimates were 2.0166 and 1.4323 for StarLife and MetLife respectively. The deductions are that the number of claims
from StarLife and MetLife is not purely independent as required by the classical Poisson process.
The Number of Claims is duration and occurrence dependent because the composition of the number of portfolios
are constantly varying (either increasing or decreasing) as a result of continuous sales of the policy.
Table 4: Criteria for assessing Goodness of Fit of the Negative Binomial Models
StarLife Criteria DF Value Value/DF
Deviance 59 69.3796 1.1759
Scaled Deviance 59 69.3796 1.1759
Person Chi-Square 59 46.7985 0.7932
Scaled Person Chi-Square 59 46.7985 0.7932
Log Likelihood 1667.5653
MetLife Criteria DF Value Value/DF
Deviance 59 71.2286 1.2073
Scaled Deviance 59 71.2286 1.2073
Person Chi-Square 59 36.4253 0.6174
Scaled Person Chi-Square 59 36.4253 0.6174
Log Likelihood 2175.5353
In effect the assumption of independence of the random process is being violated hence the inadequacy of the
Poisson model. However, the Negative Binomial Distribution produced considerably good fits in the sense that it is the
limiting form of the resulting distributions that arise in the situation of occurrence and duration dependence and is known
as Pölya - Eggenberger distribution.
Again, the number of claim process is not homogeneous as required by the Poisson distribution. The occurrence of death which drives the number of claims varies with every policy holder as a result of unobservable social, moral,
economic and health factors. Again, the Poisson distribution assumption of homogeneity is being violated and the
Negative Binomial Distribution arises as the limiting distribution and for that matter is adequate.
Bootstrap Estimates
Table 5 shows the summary statistics of the bootstrap estimate of 100 resamples from the estimated probability
distribution of the number of claims from StarLife and MetLife. A comparison between the expected monthly number of
Claims θ̂ obtained by the probability models and that obtained from the bootstrap method θ̂cshowed no significant
variation as they were set at thirteen (13) and seventeen (17) claims for StarLife and MetLife insurance company respectively.
Table 5: Summary statistics of Bootstrap replicates of Number of Claims from StarLife and MetLife
Company θ̂ θ̂* Bias θ̂c ( )ˆSE θB
StarLife 13.0500 12.9978 -0.0522 13.1022 1.9932
MetLife 16.6500 16.7274 0.0774 16.5726 1.7831
G.J.C.M.P.,Vol.3(5):41-49 (September-October, 2014) ISSN: 2319 – 7285
49
5. Conclusion and Recommendations This research only focuses on choosing between the Poisson and the Negative Binomial distribution for fitting
insurance claims and estimating the monthly expected number of claims. The Negative Binomial distribution appear to
be superior to the Poisson distribution for fitting insurance claims and therefore, provides somewhat reliable estimates for
planning, decision-making as well as estimation in insurance administration. The bootstrap estimates did not vary from
the estimates obtained by the probability models.
We therefore recommend the negative binomial distribution over the Poisson distribution for fitting insurance
claim. The bootstrap estimates should be obtained and compared with the estimates from the probability models to
authenticate the estimates. Lastly, we recommend that further work should be conducted using other models including
mixed Poisson probability models.
References Asmussen, S; Albrecher, H.(2010) Ruin Probabilities. In: Advanced series on statistical science & applied probability ; v. 14. Edition: 2nd ed. World Scientific Publishing Co.
Asmussen, S., Kliippelberg, C, 2008. Large deviations results for sub-exponential tails, with applications to insurance risk. Stochastic Process. Appl. 64, no. 1, 103-125.
Hall P. and Yao Q. (2003). “Inference in ARCH and GARCH Models with Heavy-Tailed Errors”, Econometrica 71, pp. 285-317.
Konstantinides, D., Tang, Q., Tsitsiashvili, G., 2002. Estimates for the ruin probability in the classical risk model with constant interest force in the presence of heavy tails. Insurance Math. Econom. 31, no. 3, 447-460.
Mikosch, T., Nagaev, A.V., (2007). Large deviations of heavy-tailed sums with applications in insurance. Extremes 1, no. 1, 81-110.
Miguel, J.A. and Olave, P. (1999a). “Forecast Intervals in ARCH Models: Bootstrap Versus Parametric Methods”, Applied Economics Letters 6, pp. 323-327.
Migue, J. A. and Olave, P. (1999b). “Bootstrapping Forecast Intervals in ARCH Models”, Test 8, pp. 345-364.
Lee Y. H. and Lee M. C. (2009). “Block Bootstrapping Prediction Intervals for Financial Time Series Models: Empirical Evidence from the Taiwan Stock Market”, 2009 International Symposiums on Major Academic Disciplines, Taipei.
Pascual, L., Romo, J., and Ruiz, E. (2006). “Bootstrap Prediction for Returns and Volatilities in GARCH Models”, Computational Statistics and Data Analysis 50, pp. 2293-2312.
Ridout, M.S., Hinde, J.P. and Demetrio, C.G.B. (2001) A score test for testing a zero-inflated Poisson regression model against zero-inflated negative binomial alternatives. Biometrics, 57, 219-223
Schlegel, S., 1998. Ruin probabilities in perturbed risk models. The inter play between insurance, finance and control (Aarhus, 1997). Insurance Math. Econom. 22, no. 1, 93-104.
Tang, Q. and Tsitsiashvili, G., 2003. Precise estimates for the ruin probability in finite horizon in a discrete-time model with heavy-tailed insurance and financial risks. Stochastic Process. Appl. 108 (2003), no. 2, 299-325.
Lambert, D. (1992) Zero-inflated Poisson regression, with an application to defects in manufacturing.Technometrics, 34, 1-14.
Thombs, L. A. and Schucany, W. R. 1990. “Bootstrap Prediction Intervals for Autoregression”, Journal of the American Statistical Association 85, pp. 486–492.
Yue, F. ( 2003): Semi-parametric specification tests for discrete Probability models. The Journal of Risk and Insurance, , Vol. 70, No. 1, 73-84